Automatic identification of misleading videos using a computer network

Abstract

Machine-based video classifying to identify misleading videos by training a model using a video corpus, obtaining a subject video from a content server, generating respective feature vectors of a title, a thumbnail, a description, and a content of the subject video, determining a first semantic similarities between ones of the feature vectors, determining a second semantic similarity between the title of subject video and titles of videos in the misleading video corpus in a same domain as the subject video, determining a third semantic similarity between comments of the subject video and comments of videos in the misleading video corpus in the same domain as the subject video, classifying the subject video using the model and based on the first semantic similarities, the second semantic similarity, and the third semantic similarity, and outputting the classification of the subject video to a user.

What is claimed is: 
     
       1. A method, comprising:
 training, by a computing device, a model using a video corpus; 
 obtaining, by the computing device, a subject video from a content server; 
 generating, by the computing device, respective feature vectors of a title, a thumbnail, a description, and a content of the subject video; 
 determining, by the computing device, first semantic similarities between ones of the feature vectors; 
 determining, by the computing device, a second semantic similarity between the title of subject video and titles of videos in a misleading video corpus in a same domain as the subject video; 
 determining, by the computing device, a third semantic similarity between comments of the subject video and comments of videos in the misleading video corpus in the same domain as the subject video; 
 classifying, by the computing device, the subject video using the model and based on the first semantic similarities, the second semantic similarity, and the third semantic similarity; and 
 outputting, by the computing device, the classification of the subject video to a user. 
 
     
     
       2. The method of  claim 1 , further comprising analyzing a user profile of a user that posted the subject video, wherein the classifying the subject video is based on the analyzing user profile. 
     
     
       3. The method of  claim 1 , further comprising determining an average watch time of the subject video, wherein the classifying the subject video is based on the average watch time. 
     
     
       4. The method of  claim 1 , wherein the determining the first semantic similarities between ones of the feature vectors comprises:
 determining a semantic similarity between the title and the thumbnail; 
 determining a semantic similarity between the title and the description; 
 determining a semantic similarity between the title and the content; 
 determining a semantic similarity between the thumbnail and the description; 
 determining a semantic similarity between the thumbnail and the content; and 
 determining a semantic similarity between the description and the content. 
 
     
     
       5. The method of  claim 4 , further comprising:
 assigning respective weights to ones of the first semantic similarities; 
 receiving user feedback after the classifying the subject video; and 
 adjusting at least one of the respective weights based on the user feedback. 
 
     
     
       6. The method of  claim 4 , further comprising:
 assigning respective weights to ones of the determined first semantic similarities; 
 determining a disagreement between (i) a classification based on the first semantic similarities and (ii) a classification based on the second semantic similarity or a classification based on the third semantic similarity; and 
 iteratively adjusting at least one of the respective weights based on the determined disagreement until there is agreement between the first semantic similarities, the second semantic similarity, and the third semantic similarity. 
 
     
     
       7. The method of  claim 1 , wherein:
 the video corpus comprises videos classified using predefined classes; and 
 the generating the respective feature vectors comprises:
 extracting entities from each of the title, the thumbnail, the description, and the content of the subject video; and 
 mapping the extracted entities to one or more of the predefined classes. 
 
 
     
     
       8. The method of  claim 1 , wherein the model is one of plural different models that each have different combinations of inputs, and further comprising selecting the model from the plural different models based on inputs available for the subject video. 
     
     
       9. The method of  claim 8 , further comprising:
 classifying the subject video separately using each of the plural different models; 
 comparing results of the classifying using the plural different models; and 
 iteratively adjusting one or more weights of one or more parameters used in one or more of the plural different models based on the comparing until the results of the classifying using the plural different models all agree. 
 
     
     
       10. The method of  claim 8 , further comprising re-classifying the subject video at a later time using a different one of the plural different models. 
     
     
       11. The method of  claim 1 , further comprising:
 updating the video corpus to include the subject video and the classification of the subject video; and 
 re-training the model using the updated video corpus. 
 
     
     
       12. The method of  claim 1 , wherein the classifying the subject video comprises classifying the subject video as one of: misleading; potentially misleading; and non-misleading. 
     
     
       13. A computer program product comprising one or more computer readable storage media having program instructions collectively stored on the one or more computer readable storage media, the program instructions executable to:
 obtain a subject video from a content server; 
 generate respective feature vectors of a title, a thumbnail, a description, and a content of the subject video; 
 determine semantic similarities between ones of the feature vectors; 
 classify the subject video based on a weighted sum of the semantic similarities; and 
 output the classification of the subject video to a user. 
 
     
     
       14. The computer program product of  claim 13 , wherein the determining the semantic similarities between ones of the feature vectors comprises:
 determining a semantic similarity between a title feature vector and a thumbnail feature vector; 
 determining a semantic similarity between the title feature vector and a description feature vector; 
 determining a semantic similarity between the title feature vector and a content feature vector; 
 determining a semantic similarity between the thumbnail feature vector and the description feature vector; 
 determining a semantic similarity between the thumbnail feature vector and the content feature vector; and 
 determining a semantic similarity between the description feature vector and the content feature vector. 
 
     
     
       15. The computer program product of  claim 13 , wherein:
 the program instructions are executable to determine a title semantic similarity between the title of subject video and titles of videos in a misleading video corpus in a same domain as the subject video; and 
 the classifying the subject video further based on the title semantic similarity. 
 
     
     
       16. The computer program product of  claim 13 , wherein:
 the program instructions are executable to determine a comments semantic similarity between comments of subject video and comments of videos in a misleading video corpus in a same domain as the subject video; 
 the classifying the subject video further based on the comments semantic similarity; and 
 the misleading video corpus is a subset of videos of a video corpus, wherein each video in the video corpus is tagged with one or more predefined classes, one or more predefined domains, and one or more predefined audio/visual features, and wherein each video in the misleading video corpus is additionally tagged as known misleading. 
 
     
     
       17. A system comprising:
 a processor, one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable to: 
 obtain a subject video from a content server; 
 generate respective feature vectors of a title, a thumbnail, a description, and a content of the subject video; 
 determine semantic similarities between ones of the feature vectors; 
 classify the subject video based on a weighted sum of the semantic similarities; and 
 output the classification of the subject video to a user. 
 
     
     
       18. The system of  claim 17 , wherein the determining the semantic similarities between ones of the feature vectors comprises:
 determining a semantic similarity between a title feature vector and a thumbnail feature vector; 
 determining a semantic similarity between the title feature vector and a description feature vector; 
 determining a semantic similarity between the title feature vector and a content feature vector; 
 determining a semantic similarity between the thumbnail feature vector and the description feature vector; 
 determining a semantic similarity between the thumbnail feature vector and the content feature vector; and 
 determining a semantic similarity between the description feature vector and the content feature vector. 
 
     
     
       19. The system of  claim 17 , wherein:
 the program instructions are executable to determine a title semantic similarity between the title of subject video and titles of videos in a misleading video corpus in a same domain as the subject video; and 
 the classifying the subject video further based on the title semantic similarity. 
 
     
     
       20. The system of  claim 17 , wherein:
 the thumbnail of the subject video is a selectable by a user to play the subject video in a user interface; 
 the program instructions are executable to determine a comments semantic similarity between comments of subject video and comments of videos in a misleading video corpus in a same domain as the subject video; and 
 the classifying the subject video further based on the comments semantic similarity.